Network based models and path based features for gene prioritization

Yuechang Liu, Hanghang Tong, Xie Lei, Yong Tang

Abstract: Network analysis has been shown to be an effective and cheap way to screen genes that are associated to diseases and chemicals. The identification of features that are used to order potentially related genes is key to do this job. Though many network models and structure based features have been proposed in the literature, they do not perform well enough for such gene prioritization task, especially when the heterogeneity of such networks is taken account. In this paper, a type of heterogeneous network called Generalized Bi-relational Network (GBN) is formalized. A series of path based features on GBN are defined. Though some of the features have been used in other literature, it is the first time to evaluate them in both supervised and unsupervised learning models. The experiment on real chemical-disease-gene networks shows that the features proposed in this paper gain promising performance in both supervised and unsupervised framework.